AI agents are getting hacked via prompt injections, and crypto transactions cannot be reversed. Interlock solves this by enforcing security rules in the smart contract wallet, even if the AI is compromised.
How we used TigerGraph:
Our off-chain Gateway queries a TigerGraph risk engine before signing any transaction. Using multi-hop traversals, it checks the distance between the payee and known bad actors:
· 0–1 hops: Direct link to a threat → Block.
· 2–6 hops: Indirect link (e.g., a vendor using a mixing contract) → Review (requires human approval).
· No path: Allow.
· Graph down: Review (fail-closed).
The Next.js dashboard visualizes this attack path in real-time, showing a clear node-edge graph from the Vendor (Blue) to the Contract (Orange) to the Bad Actor (Red).
The Results:
We ran a 38-scenario benchmark comparing a Naive Wallet, an LLM-Only Guard, and Interlock.
· Naive Wallet: 89% attack success.
· LLM Guard: 78% attack success.
· Interlock: 11% attack success, 0% mandate breaks, 100% legit payments completed.
The TigerGraph Experience:
TigerGraph is blazing fast at multi-hop path traversals, making it perfect for real-time fraud detection. However, running the Community Edition locally via Docker on Apple Silicon (M-series) Macs is painful—it requires x86 emulation and heavy RAM. We wish there were native ARM64 Docker builds for easier local testing.
Interlock proves that you don't have to trust the AI; you just have to trust the graph and the smart contract.
🔗 Live Demo: https://interlock-two.vercel.app
💻 GitHub: https://github.com/subratkumarpadhy4/Interlock
Top comments (0)